Summary of Xsub: Explanation-driven Adversarial Attack Against Blackbox Classifiers Via Feature Substitution, by Kiana Vu et al.
XSub: Explanation-Driven Adversarial Attack against Blackbox Classifiers via Feature Substitution
by Kiana Vu, Phung Lai, Truc Nguyen
First submitted to arxiv on: 13 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper develops a novel explanation-driven adversarial attack against black-box classifiers called XSub. XSub uses feature substitution to replace important features identified via explainable AI (XAI) with corresponding features from a “golden sample” of a different label, increasing the likelihood of misclassification. The degree of feature substitution is adjustable, balancing effectiveness and stealthiness. XSub requires O(1) queries to the prediction and explanation models, making it cost-effective. This attack can be extended to launch backdoor attacks with access to training data. Evaluation shows that XSub is effective, stealthy, and cost-effective for a wide range of AI models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary XAI has benefits in enhancing transparency and trustworthiness, but its real-world applications are limited. One challenge is that XAI can provide insights to attackers, making AI systems more vulnerable. This paper develops a new attack called XSub, which replaces important features with ones from a different label. The amount of replacement is adjustable, balancing effectiveness and stealthiness. The attack requires only a few queries and can be used for backdoor attacks if the attacker has training data. Overall, this paper shows how to make AI systems more vulnerable. |
Keywords
* Artificial intelligence * Likelihood
